v0.2.0 — settings, promotions, provider badges

Added

• Plugin settings (model-pricing namespace): refresh interval, catalog
  source URL, promotion-feed URL and tag rules — editable in settings.yaml or
  through a card on the Settings → Plugins tab; changes take effect without a
  restart (catalog fields trigger a rebuild, TTL applies to the staleness
  window immediately).
• Community promotion feed (B3): promos/<provider>.json records with a
  strict schema (expiry, verification date, attribution; either a percentage
  discount capped at 90 or a fixed per-1M price), CI-validated on every pull
  request and compiled to the feed branch on merge. The host attaches active
  promotions to matching rows; expired records never render, feed outages never
  hide prices.
• Promotion badges on the pricing table: accent badge with provenance
  tooltip (offer text, expiry, verified-by, link), a "Promotions" filter chip,
  and an active-offers count in the status line.
• Whole-provider offers: feed records without a model describe a provider
  as a whole and surface both on the provider's card in the Models settings page
  and in a "Current promotions" panel at the top of the pricing table — including
  gateways that are not in the pricing catalog at all. The feed is seeded with
  B.AI's eight documented promotions (API free tiers and discounts, verified
  against the official pricing-notices page, with a re-verification horizon).
• Provider-card badges on the Models page: each configured provider's card
  shows its lowest per-1M output price, priced-model count and active-offer
  count, straight from the catalog.
• Subscription-plan grouping: coding/token plan providers are collected
  into one group, their nominal $0 per-token prices render as "sub.", and plan
  routes are excluded from cross-provider comparison (they previously won
  "cheapest" for 39 models on a $0 price that is not a per-token rate).
• Russian language pack: the harness ships English and Chinese; this plugin
  additionally registers Russian, selected automatically for Russian browsers.
• Disk cache under the DSH home: an offline cold start serves the last real
  catalog instead of the reduced built-in snapshot.

Changed

• The "Only mine" view now follows provider topology live (llm/adapters-updated).
• HTTP responses are gzip-compressed (~8× smaller wire payload).

Install: npx -y @deepseek-ai/dsh plugin --profile web add dsh-model-pricing

dsh-model-pricing v0.1.0

What's in

Model pricing & capability board for DeepSeek Harness, mounted into Settings → Models.

• Pricing table for ~7,200 priced models across 213 providers (models.dev), per-1M-token input/output/cache prices, context, max output
• Capability tags computed from structured fields (Coding, Agentic, Vision, Long context, Open weights, Structured output) with user-configurable rules
• Cross-provider comparison: 'cheapest · Nx' and '+N% vs ' badges per model route
• Authoritative prices for routes DSH can actually invoke (pi-ai catalog overlay); >10% catalog divergences flagged
• Offline resilience: disk cache + embedded fallback snapshot; every number labeled with source and age
• Localization: English + Chinese (platform locales) plus a Russian language pack contributed by this plugin
• Runs entirely on the user's machine; anonymous read of public data sources only

Install

dsh plugin --profile web add dsh-model-pricing

Restart dsh web, open Settings → Models.